Deep Neural Network Based Channel Estimation Method for OTFS Systems

Orthogonal time frequency space (OTFS) modulation has emerged as a promising paradigm for 6G communications due to its inherent adaptability to rapidly time-varying multipath channels. However, structural complexity of delay-Doppler domain channel responses render high-precision channel estimation a persistent challenge. To this end, this paper proposes a novel deep residual attention network (DRANet) dedicated to channel estimation for OTFS systems. The model is built upon a multi-layer residual learning framework that employs deep alternating stacks of convolutional layers and residual modules to enhance nonlinear feature abstraction capability and training stability. Integrated with a channel attention mechanism, it achieves adaptive modeling of critical path gain features, thereby efficiently capturing the energy structure in the DD domain. Experimental results demonstrate that DRANet achieves superior normalized mean square error (NMSE) performance across various SNR regimes compared to conventional deep learning methods (e.g., Deep Neural Network (DNN) / Convolutional Neural Network (CNN)), while exhibiting excellent estimation accuracy and generalization capability. This fully validates its application potential in complex high-mobility scenarios.